List your company

Embodied Multi-step Decision-making and Interactive Control on ALFWorld

90.9Success RateCloud-only
46.90858.32969.7581.171May 23, 2026

Evaluation results

MethodLinks
Cloud-onlyCloud model=Qwen-Max2026.0590.911.7174.1
HeraRouting method=Dynamic routing, Backbone=Qwen2.5-0.5B2026.0586.14105.5
FrugalGPTRouting method=Dynamic routing2026.0583.78.7203.7
RouteLLMRouting method=Dynamic routing2026.0583.69156.7
EagleRouting method=Dynamic routing2026.0580.38.5178.5
Hybrid LLMRouting method=Dynamic routing2026.0579.37.9130.2
Rand. (0.5)Routing method=Stochastic step-level routing, Routing probability=0.52026.0579.18.9169.5
Rand. (0.4)Routing method=Stochastic step-level routing, Routing probability=0.42026.0577.97.2148.1
Rand. (0.3)Routing method=Stochastic step-level routing, Routing probability=0.32026.0572.16.1157.6
Device-onlyOn-device model=Qwen2.5-7B-Instruct2026.0548.60123